Research On Frequency Security Risk Assessment And Emergency Control Decision-making Of Large-scale Power Grid With Multi-infeed HVDC

It is an important means to solve the problem of reverse distribution of energy and loads in China to develop High Voltage Direct Current(HVDC)power transmission techniques to realize the optimal cross-regional allocation of energy.With the continuous construction of HVDC transmission projects,several power grids with multi-infeed HVDC have been formed in China.The equivalent inertia and spare reserve of the receiving grids decrease continuously,and the ability of adjusting frequency of the receiving grids is weakened.The risk of serious frequency deviation or even frequency security instability in receiving-end grids increases dramatically under large power disturbance such as HVDC blocking.On this background,it is helpful to support the stable operation of China’s power system to research methods about frequency security risk assessment and emergency control scheme decision-making of large-scale power grids with multi-infeed HVDC.There is uncertainty on both sides of source and loads in modern power systems,the operation modes of the power system are complicated and changeable.There are a large number of possible operation scenarios in the future,the traditional frequency security assessment methods based on a single determined scenario cannot fully reflect the future frequency security situation of the power system.It is necessary to quickly and accurately assess the frequency security of a large number of possible future scenarios,and obtain the frequency security risk indicators considering the probability information of future scenarios.Then high-risk events about frequency security can be identified,and effective emergency control schemes can be formulated in advance to ensure frequency security stability of the power system.Frequency security assessment,frequency security risk analysis and emergency control decision-making are studied in this dissertation,and the main achievements and innovations of this dissertation are as follows:(1)In the aspect of frequency security assessment,a data-driven frequency security assessment model based on Generative Adversarial Network(GAN)and Metric Learning(ML)is proposed.Firstly,the key frequency security indicators reflecting the dynamic process of frequency security after disturbance are selected as the outputs of the model,the input feature set of frequency security assessment model is constructed based on steady-state power flow information which reflects the operation characteristics of power systems and model parameter information which is related to frequency dynamics,and the key input features are obtained by dimensionality reduction;Then,the improved Wasserstein Generative Adversarial Network(WGAN)based on Wasserstein distance metric is used to learn the distribution information of historical operation scenarios of power systems,to generate operation scenarios covering typical operation modes to build the training sample set of the model,and the generated scenarios are adjusted using rejection sampling and resampling techniques,to improve the generalization ability of the frequency security assessment model;Finally,considering the inapplicability of a single machine learning model to frequency security assessment under complicated operation modes of power systems,a combined assessment model for frequency security assessment composed of multiple sub-models is constructed based on Metric Learning for Kernel Regression(MLKR)method,to learn the complicated function mapping relationship between frequency security indicators and input features in different operating scenarios.The example analysis shows that through the proposed date-driven frequency security assessment model based on GAN and ML,frequency security indicators after severe power disturbance such as HVDC blocking can be quickly and accurately assessed.(2)In the aspect of frequency security risk analysis,a risk assessment method in the way of rolling assessment for frequency security considering the uncertainty of source and loads of power systems is proposed.Firstly,Long Short Term Memory network(LSTM)is used to learn the characteristics of time sequence of source and loads,and Conditional Wasserstein Generative Adversarial Network(CWGAN)is used to model the conditional probability distribution of prediction errors of source and loads,future source and load scenarios are predicted to construct the set of future uncertain operation scenarios together with information about future plans;Then,a time-segment frequency security assessment model is built based on the multi-task learning structure of Multi-gate Mixture of Experts(MMOE),and semi-supervised learning algorithms are used to update the model based on prediction information refreshed in the rolling way,to improve the adaptability of the frequency security assessment model to the ever-changing operation modes of the power system;Finally,the frequency deviation off-limit risk is calculated and compared with the dynamic risk threshold values,to identify high-risk frequency security events triggering Under Frequency Load Shedding(UFLS)and provide them to the follow-up control part for effective emergency control scheme decision-making in advance.The example analysis shows that the proposed rolling assessment method for frequency security risk can accurately identify high-risk frequency security events considering the uncertainty of future source and loads.(3)In the aspect of emergency control scheme decision-making,a method for multi-resource emergency control scheme decision-making for power grid with multi-infeed HVDC after HVDC blocking is proposed,which includes two stages of the pre-optimization stage and online optimization stage.Firstly,the control characteristics and control cost of different control measures are analyzed,and the model of multi-resource coordinated emergency control scheme decision-making is established in order to ensure that the security indicators of the power system satisfy the threshold requirements while the control cost of the scheme is reduced as much as possible;Then,a two-stage decision-making framework is proposed,the knowledge base about emergency control scheme decision-making is established with machine learning methods in the pre-optimization stage,and the initial optimization value is provided to the subsequent online optimization of the scheme,for ensuring that the online optimization of the scheme based on accurate time-domain simulation can be completed in a short forward-looking time and the adaptability of the scheme to the actual operation mode is improved.Finally,to the high-dimensional nonlinear optimization problem of emergency control scheme decision-making,the successive linearization method is used to turn it into a mixed integer linear programming problem based on the trajectory sensitivity method to obtain the optimum control scheme.The example analysis shows that based on the proposed two-stage decision-making method of emergency control schemes,an emergency control scheme adapted to the actual operation mode of the power system can be obtained,avoiding the serious consequences that massive loads are shed passively due to the triggering of UFLS.